Research on Authorship Attribution of Article Fragments via RNNs

摘要

Most approaches for authorship attribution are based on traditional machine learning methods. Although these methods are effective, it is difficult for human to choose the features. we present a new model for authorship attribution by using Recurrent Neural Networks(RNNs) because of their powerful abilities of representation and extraction of features. We also use attention mechanism to capture the vital information in the sequence. In addition, a new method for data processing is introduced to meet the needs of the new tasks of authorship attribution. Experimental results show that the approach proposed in this paper is superior to the approach based on traditional machine learning methods.